US5257354A - System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results - Google Patents

System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results Download PDF

Info

Publication number
US5257354A
US5257354A US07/641,987 US64198791A US5257354A US 5257354 A US5257354 A US 5257354A US 64198791 A US64198791 A US 64198791A US 5257354 A US5257354 A US 5257354A
Authority
US
United States
Prior art keywords
instruction
serializing
data
instructions
decoding
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
US07/641,987
Other languages
English (en)
Inventor
Steven T. Comfort
John S. Liptay
Charles F. Webb
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
International Business Machines Corp
Original Assignee
International Business Machines Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by International Business Machines Corp filed Critical International Business Machines Corp
Priority to US07/641,987 priority Critical patent/US5257354A/en
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION reassignment INTERNATIONAL BUSINESS MACHINES CORPORATION ASSIGNMENT OF ASSIGNORS INTEREST. Assignors: COMFORT, STEVEN T., LIPTAY, JOHN S., WEBB, CHARLES F.
Priority to DE69130858T priority patent/DE69130858T2/de
Priority to EP91118905A priority patent/EP0495165B1/fr
Priority to CA002056715A priority patent/CA2056715A1/fr
Priority to JP3318815A priority patent/JPH0785222B2/ja
Priority to BR929200054A priority patent/BR9200054A/pt
Application granted granted Critical
Publication of US5257354A publication Critical patent/US5257354A/en
Anticipated expiration legal-status Critical
Expired - Fee Related legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
    • G06F9/3836—Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
    • G06F9/3861—Recovery, e.g. branch miss-prediction, exception handling

Definitions

  • the present invention relates to the processing of serializing instructions in electronic computer systems.
  • the detection is generally done by a (sometimes large) number of compare circuits, and the action to make sure that the correct value is used is carried out by special datapath circuitry, and/or by changes in the control circuitry to delay operation of some (perhaps large) portion of the machine until the needed value is available.
  • the implementation of the detection and correction logic which is needed to maintain this simple sequential view of the CPU's operation is the central problem in designing these overlap mechanisms.
  • multiprocessing Another complexity which affects the sequential view of CPU operation is multiprocessing.
  • two or more CPUs are connected to a single main storage, and operate on the contents of that storage simultaneously. This is done in order to allow greater processing power to be brought to bear on a single set of problems than could be otherwise accomplished using a single CPU.
  • multiprocessors are used with multiprogramming systems.
  • a multiprogramming system is a system of programming in which a number of separate user programs are presented for running, and a program called an operating system controls the running of the separate user programs, making sure that each one gets a fair chance to run on the CPU(s).
  • the CPUs are assigned to different user programs at any given moment.
  • the number of CPUs available increases the aggregate amount of processing power available to work on the total workload, but is not concentrated on a single user program. Since at most moments each CPU is working on a different user program which is in a different portion of storage, in many respects they are operating as completely separate systems most of the time. Nevertheless, this is not true at all times.
  • serialization point the CPU must complete all storage references which are conceptually prior to that point before doing any storage operations which conceptually follow that point. That is, at these points the CPU reverts to the simple sequential mode of operation.
  • the architecture defines that serialization points occur for all interruptions and for a certain (limited) subset of the instructions which are called serializing instructions. For most of the serializing instructions there are two serialization points, one before it begins execution, and another after it completes execution.
  • An example of such a serializing instruction in System/370 architecture is "COMPARE AND SWAP".
  • IBM CPUs implemented serialization in the simple straightforward way; that is, they actually stopped operation and waited for all prior stores to finish, before resuming operation by initiating a fetch of the next instruction. This is clearly the simplest and most straightforward way to implement this architecture, but it is also the implementation with the lowest performance. The frequency of occurrence of these events is such that this implementation was tolerable, although not entirely acceptable.
  • the architecture of a system does not define how a CPU is implemented; it only defines the results that need to be achieved. In particular, even though the architecture defines that the CPU must stop at a serialization point, and wait for stores to complete before proceeding, this is only a definition of what the correct results are. Any implementation which achieves the same results is correct.
  • This invention provides a system and method whereby the CPU can continue operation beyond a serialization point before the architecture defines that it is permissible to do so. According to the present invention, it is ascertained whether correct results are being achieved after the serializing point. If some doubt develops about the correctness of the results, the processor is returned to its pre-serialization point status and the processing is repeated.
  • correctness of results is determined by way of a monitoring mechanism.
  • This mechanism depends on the fact that the interactions between CPUs are confined to references to storage. In the disclosed mechanism the operations which are performed prior to the time that the architecture allows them, are restricted to ones do not change storage locations; that is, the CPU is not permitted to change any storage locations during this processing.
  • the needed assurance of correct operation is gained by monitoring the storage locations from which fetches are made on behalf of instructions which logically follow the serializing operation, but which are made prior to the time that fetching is allowed to resume. If those storage locations are not changed during the time between when the first such fetch is actually made from one of them, and the time that fetching is allowed to resume, then the results of the processing which was done by the CPU (based on those fetches) must be exactly the same as if all of the fetches and all of the processing was done in a single instant at the moment that the fetches became allowed. Since the latter would meet the architectural requirements, the former must also.
  • This monitoring mechanism when it finds that the storage locations were not modified, provides the needed assurance that the architectural requirements were met.
  • the monitoring mechanism finds that the storage locations may have been modified, then it causes a mechanism to be invoked which restores the CPU to an internal operating point directly following the serialization point, and causes it to resume operation from that point.
  • FIG. 1 illustrates a multiprocessor system
  • FIG. 2 illustrates the instruction buffering, and instruction decoding portion of the CPU.
  • FIG. 3 illustrates the instruction cache
  • FIG. 4 illustrates the main EXECUTION LOGIC 54 of the CPU, the CONDITIONAL PATH CONTROLS 55, and the SERIALIZATION CONTROLS 56.
  • FIG. 5 shows the DATA CACHE 73, and two mechanisms associated with it, the FETCH QUEUE 65 and the STORE QUEUE AND BUFFERS 64.
  • FIG. 6 is a flow chart of instruction fetching and setting of the SIIC bit according to a preferred embodiment of present invention.
  • FIG. 7 is a flow chart of instruction decoding and SIIC monitoring according to a preferred embodiment of the present invention.
  • FIG. 8 is a flow chart of instruction cache monitoring according to a preferred embodiment of the present invention.
  • FIG. 9 is a flow chart of the operation of the data cache controls according to a preferred embodiment of the present invention.
  • FIG. 10 is a diagram of a conditional path recovery mechanism.
  • FIG. 11 shows the format of an entry in the ACL of FIG. 10.
  • the present invention provides a system and method whereby the CPU can continue operation beyond a serialization point before the architecture defines that it is permissible to do so. According to the present invention, it is ascertained whether correct results are being achieved after the serializing point. If some doubt develops about the correctness of the results, the processor is returned to its pre-serialization point status and the processing is repeated. In the preferred embodiment, correctness of results is determined by way of a monitoring mechanism.
  • the operation of the monitoring mechanism depends on the presence of a cache in the system and the manner in which it operates, and it is therefore necessary to begin the discussion of the monitoring mechanism by describing the cache structure and operation.
  • the cache is a split cache, which is in essence two separate caches, one an instruction cache (I-Cache), and one a data cache (D-Cache). All fetches for instructions are directed to the instruction cache, and all fetches and stores for operands are directed to the data cache. During the course of operation, those storage locations which are the object of instruction fetching are loaded in the instruction cache, and those to which operand references are made are loaded in the data cache. Since these references are usually to different locations, there will be largely different data in each cache, although it is certainly possible for the same cache line to be present in both caches.
  • the data cache is one which requires exclusive control of a line when it stores into it; that is, at the time a store takes place, the line does not exist in any other cache in the system.
  • the CPU discovers that it needs to store into a storage location, the first thing it does is to examine the data cache to see if it has the line loaded, WITH exclusive control of the line. If it does not, then the address of the line is sent to all other CPUs in the system requesting that any of them which have a copy of the line invalidate it in their caches.
  • a line is loaded from storage (just for fetching) its address is broadcast in order to ensure that no other CPUs have it in their cache with exclusive control. This process of broadcasting addresses is known as cross invalidation, and only after it is complete can the CPU proceed to use a line that it has obtained.
  • a logical conclusion that can be drawn about this process is that if a cache contains a line, either with exclusive control or without it, then we are certain that no other CPU in the system is changing it. If a line leaves the cache for any reason, either as a result of being invalidated because of another CPU's request, or because of normal aging, then we are no longer certain that no other CPU is changing it.
  • the instruction cache is never stored into by the CPU, and never holds lines with exclusive control.
  • the CPU is gaining exclusive control of a line by broadcasting its address to all other CPUs in the system, its own instruction cache does not participate in the process; rather, it is allowed to retain a copy of the line.
  • the address of the line is sent to the instruction cache, and at that time the line is invalidated in the instruction cache. If that line is still needed for instruction fetching, then when the next instruction fetch is made from that line it will be reloaded from main storage, and if the CPU stores into it again, it will again be invalidated in the instruction cache.
  • the design choice is whether to permit the fetches which are made before they are architecturally permitted to consist of both instruction fetches and operand fetches, or to restrict it to instruction fetches.
  • the advantage of permitting both types of fetches to be made is that it increases the amount of processing that can be done prior to the point where fetching is permitted.
  • the disadvantage is that it is then necessary to monitor both caches, and that increases the probability that a line will be found to be removed making it necessary to go back to the instruction following the serialization point and start processing over again.
  • the desired implementation is one in which the processing of instructions normally continues smoothly past the serialization point. That is, we want an implementation in which the instructions up to and including the serialization instruction are processed with full normal overlap, and subsequent instructions are processed without any break in that overlap except that their operand fetches are held up in the fetch queue until the point at which they are architecturally allowed has been reached.
  • monitoring should begin that is, monitoring should start as soon as the fetch for the first instruction following the serialization point is made.
  • this instruction fetch occurs (most likely) before the serializing instruction reaches the instruction register; that is, before we know that the serializing instruction exists. Let us consider how can we begin monitoring before we know that we need to be monitoring.
  • This information consists of a single bit for each doubleword which indicates whether the doubleword is still in the I-cache. It is called the SIIC bit, which stands for "still in I-cache".
  • the SIIC bit is set (turned ON) during the instruction prefetching process (illustrated in FIG. 6). Whenever a doubleword of instructions is fetched from the I-cache (at 602), it is stored in the instruction buffers (I-Buffers) and its corresponding SIIC tag bit is turned ON (at 604). The SIIC bit is carried with the doubleword and is reset (turned OFF) whenever a line is removed from the instruction cache. As is conventional, the central processor will move instructions from the instructions buffers to the instruction registers (I-REGS) as I-REGS become available. As each instruction is sent from the I-BUFFERS to the I-REGS its associated buffer is made available to hold another instruction from the I-cache.
  • I-REGS instruction registers
  • Resetting of the SIIC bit is controlled by the I-cache monitoring process (illustrated in FIG. 8). Whenever a line is removed from the I-cache (at 802) all of the SIIC bits in the CPU are turned OFF (at 804). Thus when an instruction reaches the decoding logic, if its SIIC bit is ON that means that during the time since it was fetched no lines have been removed from the I-cache.
  • the instruction decoding/SIIC monitoring process is illustrated in FIG. 7.
  • instructions from the instruction buffers are gated into the instruction registers (at 702). These instructions are, in turn, decoded (at 704).
  • the decoding of further serializing instructions is inhibited (at 718)
  • data cache fetch processing is restricted (at 720)
  • a back up copy of of the CPU's register contents is made (at 708) and the SIIC monitoring process commenced (at 710).
  • the SIIC monitoring process is examined (at 714) to see if its SIIC bit is ON.
  • the operation of the CPU is different in several ways.
  • the decoding of further serializing instructions is inhibited (at 718). In other words, no further serializing instructions are allowed to decode.
  • the operation of the fetch queue is restricted (at 720). Once the instruction following the serializing instruction has been decoded (assuming the SIIC bit is on) direct monitoring of the I-Cache is commenced (at 722) and monitoring of the SIIC bit is ended.
  • the controls for inhibiting the decoding of further serializing instructions and for restricting the operation of the fetch queue will be described in more detail later.
  • any removal of a line of from the I-Cache will cause the CPU to be reset (at 716) to its status just after decoding of the serializing instruction (at 716) and to begin refetching from just after the serializing point.
  • the I-cache monitoring process is illustrated in FIG. 8. The removal of a line from the I-cache is indicated by a signal generated at block 806.
  • fetch queue processing is restricted as will now be described by reference to FIG. 9. All requests associated with the serializing instruction are sent to the fetch queue and queued up (at 904). The fetch queue is informed that all requests for the serializing instruction have been made. Then, the requests for all the instructions following the serializing instruction are sent to the fetch queue and queued up. As a result of the above-described process, the fetch queue will receive the requests as normal, but will be able to divide the requests into three groups; those requests that preceded the serializing instruction, those for the serializing instruction, and those for instructions that follow the serializing instruction.
  • the fetch queue will wait for all prior stores to be completed, and only then will it allow subsequent requests to be processed. For some serializing instructions no requests are needed, and for them the two points at which the fetch queue waits for stores to be completed blend into a single point (908). When the second (or only) of these two points is passed, then the processor operation returns to normal (at 916), and another serializing instruction can be decoded.
  • This invention can make use of whatever branch recovery mechanism is appropriate for the way in which the CPU is implemented.
  • the mechanism is provided is of the back-up register assignment list type, described in detail in U.S. Pat. No. 4,901,233 entitled "Computer System With Logic For Writing Instruction Identifying Data Into Array Control Lists For Precise Post-Branch Recoveries", which is assigned to the same assignee as the present invention.
  • FIGS. 1 through 5 An apparatus according to a preferred embodiment of the present invention will now be described by reference to FIGS. 1 through 5.
  • FIG. 1 illustrates a multiprocessor system.
  • the central piece of this system is the SYSTEM CONTROL ELEMENT 6 which connects together a number of CPUs 1,2,3, the Input/Output 7 portion of the system, and the Main Storage 8.
  • the SYSTEM CONTROL ELEMENT 6 provides all of the communication paths needed between these elements, it buffers data, addresses, and control signals as needed, and it may contain a cache.
  • Each of the CPUs is independently capable of performing all of the operations necessary to execute a program written for this system, and will normally include a cache in order to improve performance.
  • each CPU contains two caches, an I-cache to buffer instruction lines, and a D-Cache to buffer operand lines.
  • FIG. 2 illustrates the instruction buffering, and instruction decoding portion of the CPU. It includes a number of instruction buffers (I-Buffers) which hold instructions that have been fetched and are awaiting decoding. These are represented by I-Buffers 1, 2, and 3 (reference numerals 19, 20 and 21, respectively).
  • the I-Buffers 19-21 can receive their data directly from the I-cache over a bus 22, and/or can receive their data from one another as in the case of I-Buffer 3 (21) which receives its data from I-Buffer 2 (20) over a bus 23.
  • the I-Buffers 19-21 are preferably of a size which is equal to the amount of data which is delivered from the I-cache in a single cycle, which is commonly a doubleword in a CPU that implements the System/370 architecture.
  • the ingating to the I-Buffers 19-21 is controlled by a set of controls 24 which keep track of what data is in the I-buffers, request additional data from the I-cache as needed, remember what is on the way, and keep track of what instructions are being taken by the I-register for decoding.
  • the controls 24 control ingating to the I-buffers with a set of control signals 29,30,31.
  • the controls 24 communicate with the I-cache over a set of control lines 25 which are used to make requests, send the addresses from which data is wanted, and receive information about the progress of requests.
  • the controls 24 also control a set of gates 32 which select data that is to be sent to the I-REG 33, and they communicate with the DECODING AND I-REG CONTROLS 34 over a set of signals 27.
  • the instruction register (I-REG) 33 which receives instructions from the gating logic may contain one or more instructions which are presented to the DECODING AND I-REG CONTROLS 34 so that they may be examined and passed on to the appropriate portions of the execution logic for execution. All of this logic is structured in the way that is appropriate for the needs of the particular CPU that is being designed.
  • SIIC bit 35,36,37,38 For each I-buffer, and I-register, there is an accompanying bit 35,36,37,38 called an SIIC bit. This bit is controlled by the same signals 29,30,31,39 which control ingating to the I-buffers and the I-REG. Whenever data from the I-cache is ingated into one of the I-buffers, such as I-BUFFER 1 (19), then the associated SIIC bit 35 is set to ONE. Whenever data is ingated from one I-buffer to another, such as from I-BUFFER 2 (20) to I-BUFFER 3 (21), then the SIIC bit associated with the source I-buffer 36 is gated to the SIIC bit associated with the destination I-buffer 37.
  • the SIIC bit associated with the I-buffer from which the first byte of data came is gated to the SIIC bit associated with the I-REG 38.
  • the DECODING AND I-REG CONTROLS 34 contains the logic which examines the instruction(s) in the I-REG 33, determines whether the instruction can be passed to the EXECUTION LOGIC [54 in FIG. 4], and decodes a number of signals which are helpful in moving the instruction to the correct portion of the execution logic. There are times when an instruction is decoded, and it is not clear what instruction is correct to decode next. One example of a situation like this is when a branch is encountered. Until the EXECUTION LOGIC 54 determines whether the branch is taken, it is not known for sure whether the next instruction to be executed is the sequentially next one in storage, or the one at the target address.
  • the DECODING AND I-REG CONTROLS 34 makes a guess about which direction the branch will go.
  • the preferred mechanism for making this guess is a conventional branch history table (an apparatus that makes a branch direction guess based on data recorded about prior operation of the CPU) although this is not a requirement of the present invention.
  • Alternative mechanisms such as an apparatus for making a guess based on simple examination of the instruction, can be substituted for the branch history table. If the decision is to continue with the sequential instructions, then the DECODING AND I-REG CONTROLS 34 continues gating instructions from the GATING logic 32 into the I-REG 33 and processing them.
  • the DECODING AND I-REG CONTROLS 34 signals the I-BUFFER Controls 24 to clear the I-buffers and start fetching instructions over again, this time from the target address.
  • all of those instructions beyond the branch are considered to be conditional; that is, depending on the outcome of the branch they may really end up being executed, or it may be necessary to undo them.
  • there are some CONDITIONAL PATH CONTROLS (55 in FIG. 4) which the DECODING AND I-REG CONTROLS 34 communicates with over signals 58.
  • the CONDITIONAL PATH CONTROLS 55 are described in greater detail below, but at this point, it would be noted that the controls are preferably of a type that allows there to be multiple conditional paths past multiple unresolved branches.
  • this invention can use a conditional path mechanism designed for branches (of any of a number of types known in the prior art), for a different operation, serializing instructions.
  • What serializing instructions share in common with branches is an uncertainty about what instruction is to be executed next. In the case of branches, this uncertainty derives from not knowing for sure where the next instruction should be fetched from. In the case of serializing instructions, we know for certain where it will be fetched from, but we do not know whether some other processor may change it before we reach the serializing point.
  • the DECODING AND I-REG CONTROLS 34 communicate with a set of SERIALIZATION CONTROLS (56 in FIG. 4) over a set of signals 59 both as part of process of decoding the instruction, and to initialize those controls for processing the instruction.
  • FIG. 3 illustrates the instruction cache. This cache contains two arrays, a DATA ARRAY 41 and an ADDRESS ARRAY 42.
  • the DATA ARRAY 41 contains lines of data from main storage, and for each line 43 in the DATA ARRAY, the ADDRESS ARRAY 42 contains an address 44 which identifies where in main storage the line came from, and control bits to record the status of the line.
  • the I-CACHE Controls 47 control the movement of data to and from the arrays 41,42, and the placement of data in those arrays.
  • the I-CACHE Controls 47 use a set of signal lines 25 to communicate with the I-BUFFER Controls (24 in FIG. 2).
  • the signals 25 provide information about the addresses of locations from storage that need to be fetched, about how those fetches are being processed, and when data will be returned. Data is sent f&om the instruction cache over a bus 22 from the DATA ARRAY 41 to the instruction buffers (19,20 in FIG. 2).
  • the instruction cache communicates with the SYSTEM CONTROL ELEMENT [6 in FIG. 1] using some of the signals 9 between the CPU and the SYSTEM CONTROL ELEMENT 6.
  • These signals 9 include both control signals that go to and from the I-CACHE Controls 47 to coordinate the operation of the two parts of the system, and data signals which go to the DATA ARRAY 41 for the actual return of data.
  • the I-CACHE Controls 47 When lines of data are moved into the instruction cache, the I-CACHE Controls 47 also communicate with the ADDRESS ARRAY 42 over signals 46 to determine where in the cache the line should be placed, and to record its presence, and with DATA ARRAY 41 over signals 45 to make sure that the data is placed in the correct place in the DATA ARRAY.
  • signals 9 from the SYSTEM CONTROL ELEMENT 6 are signals which indicate when some other processor in the system wants exclusive control of a particular line of main storage, or when it wants non-exclusive control and needs to make sure that this CPU does not have exclusive control. Those signals come to the Cross Interrogate Controls 48 which are associated with the instruction cache. When these controls receive such a signal they determine whether that line is in the instruction cache, and whether it is necessary to remove it. To carry this out, they have signals 49 to communicate with the ADDRESS ARRAY 42, and other signals 50 to communicate with the I-CACHE Controls 47.
  • Another feature provided in the preferred embodiment of the present invention is a signal 87 from the STORE QUEUE (64 in FIG. 5) which comes to the Cross Interrogate Controls 48, and a signal 88 generated by the Cross Interrogate Controls 48.
  • the incoming signal 87 indicates that the FETCH QUEUE 65 has completed all requests prior to and associated with the serializing operation, and that all store operations pending in the STORE QUEUE AND BUFFERS (64 in FIG. 5) have been processed. This means that the CPU has reached the point where the architecture permits subsequent fetching to begin. However, the CPU cannot return to normal operation until any cross interrogate requests which have come in prior to this point have been processed. Therefore, the signal 87 comes to the Cross Interrogate Controls 48, and then when there are no prior cross interrogate operations pending, the Cross Interrogate Controls 48 generate another signal 88 which notifies the rest of the CPU that normal operation may be resumed.
  • FIG. 4 illustrates the main EXECUTION LOGIC 54 of the CPU, the CONDITIONAL PATH CONTROLS 55, and the SERIALIZATION CONTROLS 56.
  • the EXECUTION LOGIC 54 contains all of the logic needed to produce the results called for by the instructions, and contains most or all of the registers needed to hold the status of the processor, such as the General Registers, the Floating Point Registers, the Control Registers, and the Program Status Word in a processor implementing the System/370 architecture. Its principal interfaces are to receive instructions from the DECODING AND I-REG CONTROLS (34 in FIG. 2) over a set of signals 57, and to make fetch and store requests by communicating with the FETCH QUEUE (65 in FIG. 5) over a set of signals 68, and with the STORE QUEUE AND BUFFERS (64 in FIG. 5) over a set of signals 69.
  • the EXECUTION LOGIC 54 works can take many different forms while still being usable with the present invention. Nevertheless, it needs the ability to deal with conditional instruction paths.
  • the EXECUTION LOGIC is of the type that contains multiple execution elements performing different operations at the same time, and possibly in an order different from the order in which those operations were called for by the program.
  • An example of such a execution logic is described in U.S. Pat. No. 4,901,233.
  • the CONDITIONAL PATH CONTROLS 55 can also take various different forms, the particular form chosen depending on the structure of the EXECUTION LOGIC 54 and of the rest of the processor. There are several examples in the prior art of structures that can be used.
  • the IBM System/360 Model 91 CPU had multiple execution elements and could execute instructions out of their logical sequence. In that processor, instructions beyond a conditional branch could be decoded and sent to the execution elements before the branch was resolved. All actions related to such instructions carried with them a conditional bit which identified them as conditional. If the branch did not go in the direction guessed, then all operations which are tagged as conditional are discontinued. This CPU only allowed decoding to proceed past a single unresolved branch.
  • the IBM System/360 Model 85 CPU executed all instructions in their logical sequence, and it allowed instructions beyond only a single conditional branch to be decoded. Because all instructions were executed in sequence, when a branch executed (and its direction was resolved), all of the instructions in the CPU were ones that were beyond the branch, and were therefore conditional. Therefore, in the event that a branch guess was wrong all operations going on in the processor needed to be discarded. That is, unlike the Model 91, there was no need to explicitly tag operations as conditional because the sequential execution of the processor made that implicit.
  • the IBM 3033 CPU executed all instructions in their logical sequence just as the Model 85 did, but it allowed instructions beyond two conditional branches to be decoded. As with the Model 85, there was no need to explicitly tag instructions as conditional because when a branch was executed (and its direction resolved) all of the operations left in progress were ones which logically followed the branch and were therefore conditional on it.
  • U.S. Pat. No. 4,901,233 describes a mechanism for effecting precise recoveries after a guessed wrong branch in a CPU which allows instructions to be executed out of sequence, which allows the execution of those instructions to proceed to the point of computing their final results and placing them in their final location in internal registers, and which allows execution to proceed past an arbitrary number of unresolved branches.
  • the CONDITIONAL PATH CONTROLS 55 are signalled by the DECODING AND I-REG CONTROLS (34 in FIG. 2) and start up a new set of controls, called a conditional path, to control the instructions along that path.
  • the CONDITIONAL PATH CONTROLS 55 provide for one or more conditional paths, and when all of them are in use the CONDITIONAL PATH CONTROLS 55 signal the DECODING AND I-REG CONTROLS (34 in FIG. 2) to not decode any more such instructions.
  • the controls for each conditional path receive the necessary inputs to determine when the instruction which caused it to be turned on is determined to have been guessed correctly or not. When this determination is made, the conditional path signals all necessary parts of the CPU that the guess was correct or wrong. This signaling is included within the signals 58 to the DECODING AND I-REG CONTROLS 34, the signals 61 to the EXECUTION LOGIC 54, the signals 71 to the SERIALIZATION LOGIC 56, the signals 62 to the STORE QUEUE AND BUFFERS (64 in FIG. 5), and the signals 63 to the FETCH QUEUE (65 in FIG. 5).
  • the various parts of the CPU take the necessary steps to adjust their operations; that is, if the guess was correct then the operations which were part of that conditional path are no longer considered conditional (as part of that path, but may still be part of another conditional path), and if the guess was wrong then they are cancelled.
  • Logic in all of the parts of the CPU that carry out actions along conditional paths is designed in such a way as to be able to carry out these operations. This logic is described in greater detail below and in U.S. Pat. No. 4,901,233. Elements from this reference are shown in FIGS.
  • a Decode Register Assignment List (DRAL) 100 is connected to incoming and outgoing status and control lines.
  • Logic unit 101 is also connected to incoming and outgoing status and control lines for monitoring and controlling the contents of the DRAL.
  • the DRAL is used when instructions are decoded to translate architected logical register assignments (e.g. GPRs named within instructions) into physical register assignments.
  • the DRAL contains one position for each logical register, and that position contains the number of the physical register that has been most recently assigned to receive value for that logical register.
  • each instruction As each instruction is decoded, the logical registers it references are looked up in the DRAL to determine what physical registers are assigned to the logical registers, and as new physical registers are assigned to receive results, the DRAL is updated to reflect these assignments. In this way, each instruction which uses a physical register is directed by the DRAL to find the physical register which is assigned to the most recent instruction to reference that logical register.
  • Back-up Register Assignment Lists 102, 104 and 106 are connected to receive the entire contents of the DRAL 100 at particular points of operation.
  • the use of one, two or three BRALs allows the processing of one, two or three conditional paths respectively, without waiting.
  • a conditional path can be created through the decoding of a conditional branch instruction or a serializing instruction.
  • Each BRAL has the same structure as the DRAL, and is connected to it in such a way that in one cycle the entire contents of the DRAL may be copied into the BRAL, or vice versa. These transfers are controlled by logic unit 101. It is used, for example, when a serializing instruction is encountered by the decoder to save the contents of the DRAL in case the instructions conditionally executed beyond the serializing instructions can not complete due to a line being removed from the I-cache.
  • conditional path causing instructions The need for creation of a conditional path can be created by decoding of either a serializing instruction or a conditional branch instruction. These two types of instructions will be referred to collectively as "conditional path causing instructions".
  • One difference, however, between serializing instructions and conditional branch instructions is that after a serializing instruction is decoded, the decoding of further serializing instructions is inhibited until after the first decoded serializing instruction has completed execution.
  • conditional branches however, have no such limitation on their decoding.
  • more than one conditional path can be active, having been caused by the decoding of a plurality of conditional branches or by decoding of a serializing instruction combined with one or more conditional branches.
  • An Array Control List (ACL) 110 is connected to receive status information and send control information from the physical registers and the rest of the computer system.
  • Logic unit 101 controls the contents of ACL 110 and coordinates the operation of the ACL and the DRAL.
  • ACL register which remembers status information related to that physical register.
  • the ACL includes one entry for each physical register, each entry, in this embodiment, consisting of 14 bits divided into five fields as shown in FIG. 11: CTL, ABC, IID, PRV and REG.
  • CTL is a control field which defines the overall status of the corresponding physical register. It may have the following values:
  • the IID field is the instruction identity number assigned to the instruction for which this physical register will receive a result.
  • the ABC field is nonzero if this instruction has been issued conditionally and "000" otherwise.
  • the REG field is the number of the logical register for which this physical register is receiving a result
  • the PRV field is the number of the physical register which holds the old value of that logical register. In the special case of compare instructions which do not change a logical register, a physical register is assigned anyway, the REG field is irrelevant, and the PRV field contains the number of this physical register.
  • the ABC field is used to identify the physical registers which are associated with instructions that are decoded beyond a conditional branch or serializing instruction. It is not necessary to implement such a field in order to make the invention work since this information could be determined by examining the IIDs. However, this field is implemented in this embodiment because it is more economical and faster to have a field devoted to remembering this one piece of information, than to determine it at periodic intervals by performing a test of the IIDs.
  • the ABC field has three bits which are initially set to be "000".
  • the first BRAL is loaded with the DRAL contents and the ABC field is changed to "100" to indicate loading of the BRAL and the presence of an uncompleted serializing instruction.
  • the second BRAL is loaded with the contents of the DRAL at that time and the ABC field is changed to "110".
  • the third BRAL receives the DRAL contents at that time and the ABC field becomes "111".
  • Each bit in the ABC field is independently associated with a particular BRAL to show that the BRAL is active and storing particular DRAL contents for an unresolved branch or a serializing instruction.
  • Logic unit 101 controls these functions.
  • the DECODING AND I-REG CONTROLS 34 signal the SERIALIZATION CONTROLS 56, and if a serialization operation is already in progress, then decoding of the instruction is delayed.
  • both the SERIALIZATION CONTROLS 56 and the CONDITIONAL PATH CONTROLS 55 recognize that and start operations to control the instruction.
  • the CONDITIONAL PATH CONTROLS 55 start up a conditional path for the following instructions so that if necessary they can be cancelled and control returned to the instruction which follows the serialization operation.
  • the SERIALIZATION CONTROLS 56 monitor the SIIC-I bit 38 and look at whether it is ON for the instruction following the serializing instruction.
  • the controls 56 signal the DECODING AND I-REG CONTROLS 34 via one of their communication signals 59 to not decode the next instruction, and instead to signal the I-BUFFER CONTROLS to clear all of the prefetched instructions in the I-buffers 19,20,21 and the I-REG 33, and refetch them.
  • the CONDITIONAL PATH CONTROLS 55 receive signals (part of 61) from the EXECUTION LOGIC 54 which provide it with the necessary information to determine whether branch instructions have been guessed correctly or incorrectly.
  • the CONDITIONAL PATH CONTROLS 55 also receive two signals from elsewhere in the CPU to enable them to determine when action needs to be taken for a serializing instruction. The first of these signals 40 comes from the instruction cache and indicates that a line has been removed from the instruction cache; this signal was described earlier. When this signal 40 comes on, the conditional path which is for the serializing instruction brings on a path wrong signal to the entire processor to cause all instructions along that path to be cancelled, and to cause decoding to return to the address of the instruction following the serialization operation.
  • the second of these signals 88 comes to the CONDITIONAL PATH CONTROLS 55 from the CROSS INTERROGATE CONTROLS (48 in FIG. 3) and indicates that the CPU has reached the time at which it is permitted to perform operations past the serializing point according to the architecture. Therefore, this signal 88 causes the conditional path for the serializing operation to signal that the path is correct, thereby removing the conditional status of the instructions along that path, and making the conditional path available for use by another branch or serializing operation. This is the same action as is taken for a correctly guessed branch.
  • the SERIALIZATION CONTROLS 56 also watch this signal 88, and when it comes on they clear themselves to allow another serializing operation to take place.
  • FIG. 5 shows the DATA CACHE 73, and two mechanisms associated with it, the FETCH QUEUE 65 and the STORE QUEUE AND BUFFERS 64.
  • the FETCH QUEUE 65 receives signals 68 from the EXECUTION LOGIC (54 in FIG. 4) which request that certain operations be performed by the DATA CACHE 73. It records these requests in the order in which they were received until the appropriate time to do them has been reached, and then performs them in the same order that they were received. Since it is possible for requests that were generated by conditional instructions to be sent to the FETCH QUEUE 65, it contains information 66 about conditional paths, receives signals 63 from the CONDITIONAL PATH CONTROLS (55 in FIG.
  • the DATA CACHE 73 is structured similarly to the instruction cache described above, although a comparable degree of detail is not provided. It serves to remember lines of data which have recently been referenced as operands, either for fetching or storing. The DATA CACHE 73 receives fetch requests over the signals 75 from the FETCH QUEUE 65, and returns data over a bus 74 to the EXECUTION LOGIC [54 in FIG. 4].
  • the DATA CACHE 73 also receives store pretest requests from the FETCH QUEUE 65 over the same set of signals 75.
  • a store pretest request occurs in anticipation of a store to an address, and is meant to determine whether the required line is in the cache with exclusive control, and whether there are any access exceptions associated with storing into the line. If the line is not in the cache, or is there but not with exclusive control, then the DATA CACHE 73 signals 9 the SYSTEM CONTROL ELEMENT [6 in FIG. 1]that it needs exclusive control of the line. This initiates the cross interrogate process described previously.
  • STORE QUEUE AND BUFFERS 64 over some signals 76.
  • the STORE QUEUE AND BUFFERS 64 serve to remember data that is to be stored, and information about that data such as the address at which it is stored in MAIN STORAGE (8 in FIG. 1), and the position in the DATA CACHE 73 at which the line is located.
  • the STORE QUEUE AND BUFFERS 64 receive data from the EXECUTION LOGIC 54 over some signals 69, and addresses for storing the data from the FETCH QUEUE 65.
  • the timing relationship in which this information is received, the structure of the information, and the manner in which it is handled by the STORE QUEUE AND BUFFERS 64 depend on details of how the CPU operates and are not germane to this disclosure.
  • conditional instructions Since it is possible for data that is generated by conditional instructions to be sent to the STORE QUEUE AND BUFFERS 64, they contain information about conditional paths, receive signals 62 from the CONDITIONAL PATH CONTROLS (55 in FIG. 4), and have the ability to cancel those stores if operation has proceeded along a wrong conditional path.
  • a store is held in the STORE QUEUE AND BUFFERS 64 until the operation it is associated with reaches completion; that is, a point has been reached at which it is certain that that (and all prior) operations are to be performed. At that time, the store is released for storing into the DATA CACHE [73]using some signals 77, and for sending to the SYSTEM CONTROL ELEMENT (6 in FIG. 1) over the signals 9 which connect the CPU to it.
  • the FETCH QUEUE 65 receives as part of its signals 68 from the EXECUTION LOGIC 54 an indication of where the serialization points are. For most serializing instructions, two points are identified, the first point being between the requests which precede those for the serializing instruction and the request which are for it, and a second point being between the requests which are for the serializing instruction and those which follow it. Some serializing instructions do not make any requests to the fetch queue, and therefore only have a single serializing point.
  • FETCH QUEUE 65 When FETCH QUEUE 65 reaches the second (or only) serializing point associated with the serializing operation, it signals the STORE QUEUE AND BUFFERS 64 (part of 76). Then, when the STORE QUEUE AND BUFFERS 64 finish processing all stores, it generates a signal 87 which is sent to the CROSS INTERROGATE CONTROLS (48 in FIG. 3) to indicate that fact.
  • the action of the CROSS INTERROGATE CONTROLS 48 is to wait until any pending cross interrogate operations have been processed, and then generate a signal 88 which is sent to the SERIALIZATION CONTROLS 56 and CONDITIONAL PATH CONTROLS 55 which informs them that the point has been reached when the CPU is allowed to proceed beyond the architected serialization point.
  • This signal 88 its significance, and the actions taken as a result of it, have been described above.

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Advance Control (AREA)
  • Memory System Of A Hierarchy Structure (AREA)
  • Hardware Redundancy (AREA)
US07/641,987 1991-01-16 1991-01-16 System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results Expired - Fee Related US5257354A (en)

Priority Applications (6)

Application Number Priority Date Filing Date Title
US07/641,987 US5257354A (en) 1991-01-16 1991-01-16 System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results
DE69130858T DE69130858T2 (de) 1991-01-16 1991-11-06 Überlappende Serienverarbeitung
EP91118905A EP0495165B1 (fr) 1991-01-16 1991-11-06 Sérialisation à chévauchement
CA002056715A CA2056715A1 (fr) 1991-01-16 1991-11-29 Serialisation a chevauchement
JP3318815A JPH0785222B2 (ja) 1991-01-16 1991-12-03 データ処理装置
BR929200054A BR9200054A (pt) 1991-01-16 1992-01-10 Aparelho para serializacao sobreposta e seu metodo de processamento

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US07/641,987 US5257354A (en) 1991-01-16 1991-01-16 System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results

Publications (1)

Publication Number Publication Date
US5257354A true US5257354A (en) 1993-10-26

Family

ID=24574686

Family Applications (1)

Application Number Title Priority Date Filing Date
US07/641,987 Expired - Fee Related US5257354A (en) 1991-01-16 1991-01-16 System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results

Country Status (6)

Country Link
US (1) US5257354A (fr)
EP (1) EP0495165B1 (fr)
JP (1) JPH0785222B2 (fr)
BR (1) BR9200054A (fr)
CA (1) CA2056715A1 (fr)
DE (1) DE69130858T2 (fr)

Cited By (28)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5488706A (en) * 1992-12-18 1996-01-30 Amdahl Corporation Retry request system in a pipeline data processing system where each requesting unit preserves the order of requests
US5613080A (en) * 1993-09-20 1997-03-18 International Business Machines Corporation Multiple execution unit dispatch with instruction shifting between first and second instruction buffers based upon data dependency
US5630157A (en) * 1991-06-13 1997-05-13 International Business Machines Corporation Computer organization for multiple and out-of-order execution of condition code testing and setting instructions
US5691920A (en) * 1995-10-02 1997-11-25 International Business Machines Corporation Method and system for performance monitoring of dispatch unit efficiency in a processing system
US5729726A (en) * 1995-10-02 1998-03-17 International Business Machines Corporation Method and system for performance monitoring efficiency of branch unit operation in a processing system
US5748855A (en) * 1995-10-02 1998-05-05 Iinternational Business Machines Corporation Method and system for performance monitoring of misaligned memory accesses in a processing system
US5751945A (en) * 1995-10-02 1998-05-12 International Business Machines Corporation Method and system for performance monitoring stalls to identify pipeline bottlenecks and stalls in a processing system
US5752062A (en) * 1995-10-02 1998-05-12 International Business Machines Corporation Method and system for performance monitoring through monitoring an order of processor events during execution in a processing system
US5797019A (en) * 1995-10-02 1998-08-18 International Business Machines Corporation Method and system for performance monitoring time lengths of disabled interrupts in a processing system
US5832258A (en) * 1993-09-27 1998-11-03 Hitachi America, Ltd. Digital signal processor and associated method for conditional data operation with no condition code update
US5949971A (en) * 1995-10-02 1999-09-07 International Business Machines Corporation Method and system for performance monitoring through identification of frequency and length of time of execution of serialization instructions in a processing system
CN1046167C (zh) * 1994-07-21 1999-11-03 张立平 稳定计算机系统输出状态的抗干扰方法
US6032244A (en) * 1993-01-04 2000-02-29 Cornell Research Foundation, Inc. Multiple issue static speculative instruction scheduling with path tag and precise interrupt handling
US6041167A (en) * 1994-03-31 2000-03-21 International Business Machines Corporation Method and system for reordering instructions after dispatch in a processing system
US6088792A (en) * 1998-04-30 2000-07-11 International Business Machines Corporation Avoiding processor serialization after an S/390 SPKA instruction
US6088791A (en) * 1998-04-30 2000-07-11 International Business Machines Corporation Computer processor system for implementing the ESA/390 STOSM and STNSM instructions without serialization or artificially extending processor execution time
US6618851B1 (en) * 1999-08-31 2003-09-09 Autodesk, Inc. Method and apparatus for state-reversion
US20040088496A1 (en) * 2002-11-05 2004-05-06 Newisys, Inc. A Delaware Corporation Cache coherence directory eviction mechanisms in multiprocessor systems
US20040088494A1 (en) * 2002-11-05 2004-05-06 Newisys, Inc. A Delaware Coporation Cache coherence directory eviction mechanisms in multiprocessor systems
US20040088495A1 (en) * 2002-11-05 2004-05-06 Newisys, Inc., A Delaware Corporation Cache coherence directory eviction mechanisms in multiprocessor systems
WO2004068361A1 (fr) * 2003-01-27 2004-08-12 Fujitsu Limited Dispositif de commande de memorisation, dispositif de commande de cache de donnees, unite centrale, procede de commande de dispositif de memorisation, procede de commande de cache de donnees et procede de commande de cache
US20050210204A1 (en) * 2003-01-27 2005-09-22 Fujitsu Limited Memory control device, data cache control device, central processing device, storage device control method, data cache control method, and cache control method
US20130046961A1 (en) * 2011-08-15 2013-02-21 Alexander Rabinovitch Speculative memory write in a pipelined processor
CN103728897A (zh) * 2012-10-12 2014-04-16 深圳市金正方科技股份有限公司 一种基于软硬件结合的抗干扰方法及系统
US20140281776A1 (en) * 2013-03-14 2014-09-18 Teradyne, Inc. Method and apparatus for device testing using multiple processing paths
US9558116B2 (en) 2003-04-09 2017-01-31 Intellectual Ventures Holding 81 Llc Coherence de-coupling buffer
US10139449B2 (en) 2016-01-26 2018-11-27 Teradyne, Inc. Automatic test system with focused test hardware
US10907426B2 (en) 2018-10-15 2021-02-02 H. Udo Zeidler Apparatus and method for early kick detection and loss of drilling mud in oilwell drilling operations

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0529303A3 (en) * 1991-08-29 1993-09-22 International Business Machines Corporation Checkpoint synchronization with instruction overlap enabled
JPH06168263A (ja) * 1992-11-30 1994-06-14 Fujitsu Ltd ベクトル処理装置
US5465336A (en) * 1994-06-30 1995-11-07 International Business Machines Corporation Fetch and store buffer that enables out-of-order execution of memory instructions in a data processing system

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3736566A (en) * 1971-08-18 1973-05-29 Ibm Central processing unit with hardware controlled checkpoint and retry facilities
US4760520A (en) * 1984-10-31 1988-07-26 Hitachi, Ltd. Data processor capable of executing instructions under prediction
US4777594A (en) * 1983-07-11 1988-10-11 Prime Computer, Inc. Data processing apparatus and method employing instruction flow prediction
US4901233A (en) * 1987-07-20 1990-02-13 International Business Machines Corporation Computer system with logic for writing instruction identifying data into array control lists for precise post-branch recoveries
US5119483A (en) * 1988-07-20 1992-06-02 Digital Equipment Corporation Application of state silos for recovery from memory management exceptions
US5136696A (en) * 1988-06-27 1992-08-04 Prime Computer, Inc. High-performance pipelined central processor for predicting the occurrence of executing single-cycle instructions and multicycle instructions
US5142634A (en) * 1989-02-03 1992-08-25 Digital Equipment Corporation Branch prediction

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH06103494B2 (ja) * 1986-11-18 1994-12-14 株式会社日立製作所 ベクトル処理装置の制御方式
US4991090A (en) * 1987-05-18 1991-02-05 International Business Machines Corporation Posting out-of-sequence fetches

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3736566A (en) * 1971-08-18 1973-05-29 Ibm Central processing unit with hardware controlled checkpoint and retry facilities
US4777594A (en) * 1983-07-11 1988-10-11 Prime Computer, Inc. Data processing apparatus and method employing instruction flow prediction
US4760520A (en) * 1984-10-31 1988-07-26 Hitachi, Ltd. Data processor capable of executing instructions under prediction
US4901233A (en) * 1987-07-20 1990-02-13 International Business Machines Corporation Computer system with logic for writing instruction identifying data into array control lists for precise post-branch recoveries
US5136696A (en) * 1988-06-27 1992-08-04 Prime Computer, Inc. High-performance pipelined central processor for predicting the occurrence of executing single-cycle instructions and multicycle instructions
US5119483A (en) * 1988-07-20 1992-06-02 Digital Equipment Corporation Application of state silos for recovery from memory management exceptions
US5142634A (en) * 1989-02-03 1992-08-25 Digital Equipment Corporation Branch prediction

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
IBM Technical Disclosure Bulletin, vol. 25, No. 11B, Apr. 1983, "Shared Cache In A Checkpoint Environment" by J. A. Weiss et al.
IBM Technical Disclosure Bulletin, vol. 25, No. 11B, Apr. 1983, Shared Cache In A Checkpoint Environment by J. A. Weiss et al. *

Cited By (38)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5630157A (en) * 1991-06-13 1997-05-13 International Business Machines Corporation Computer organization for multiple and out-of-order execution of condition code testing and setting instructions
US5983335A (en) * 1991-06-13 1999-11-09 International Business Machines Corporation Computer system having organization for multiple condition code setting and for testing instruction out-of-order
US5881308A (en) * 1991-06-13 1999-03-09 International Business Machines Corporation Computer organization for multiple and out-of-order execution of condition code testing and setting instructions out-of-order
US5488706A (en) * 1992-12-18 1996-01-30 Amdahl Corporation Retry request system in a pipeline data processing system where each requesting unit preserves the order of requests
US6032244A (en) * 1993-01-04 2000-02-29 Cornell Research Foundation, Inc. Multiple issue static speculative instruction scheduling with path tag and precise interrupt handling
US5613080A (en) * 1993-09-20 1997-03-18 International Business Machines Corporation Multiple execution unit dispatch with instruction shifting between first and second instruction buffers based upon data dependency
US5832258A (en) * 1993-09-27 1998-11-03 Hitachi America, Ltd. Digital signal processor and associated method for conditional data operation with no condition code update
US6041167A (en) * 1994-03-31 2000-03-21 International Business Machines Corporation Method and system for reordering instructions after dispatch in a processing system
CN1046167C (zh) * 1994-07-21 1999-11-03 张立平 稳定计算机系统输出状态的抗干扰方法
US5729726A (en) * 1995-10-02 1998-03-17 International Business Machines Corporation Method and system for performance monitoring efficiency of branch unit operation in a processing system
US5797019A (en) * 1995-10-02 1998-08-18 International Business Machines Corporation Method and system for performance monitoring time lengths of disabled interrupts in a processing system
US5949971A (en) * 1995-10-02 1999-09-07 International Business Machines Corporation Method and system for performance monitoring through identification of frequency and length of time of execution of serialization instructions in a processing system
US5752062A (en) * 1995-10-02 1998-05-12 International Business Machines Corporation Method and system for performance monitoring through monitoring an order of processor events during execution in a processing system
US5751945A (en) * 1995-10-02 1998-05-12 International Business Machines Corporation Method and system for performance monitoring stalls to identify pipeline bottlenecks and stalls in a processing system
US5748855A (en) * 1995-10-02 1998-05-05 Iinternational Business Machines Corporation Method and system for performance monitoring of misaligned memory accesses in a processing system
US5691920A (en) * 1995-10-02 1997-11-25 International Business Machines Corporation Method and system for performance monitoring of dispatch unit efficiency in a processing system
US6088792A (en) * 1998-04-30 2000-07-11 International Business Machines Corporation Avoiding processor serialization after an S/390 SPKA instruction
US6088791A (en) * 1998-04-30 2000-07-11 International Business Machines Corporation Computer processor system for implementing the ESA/390 STOSM and STNSM instructions without serialization or artificially extending processor execution time
US6618851B1 (en) * 1999-08-31 2003-09-09 Autodesk, Inc. Method and apparatus for state-reversion
US20040088494A1 (en) * 2002-11-05 2004-05-06 Newisys, Inc. A Delaware Coporation Cache coherence directory eviction mechanisms in multiprocessor systems
US20040088495A1 (en) * 2002-11-05 2004-05-06 Newisys, Inc., A Delaware Corporation Cache coherence directory eviction mechanisms in multiprocessor systems
US6920532B2 (en) 2002-11-05 2005-07-19 Newisys, Inc. Cache coherence directory eviction mechanisms for modified copies of memory lines in multiprocessor systems
US6925536B2 (en) 2002-11-05 2005-08-02 Newisys, Inc. Cache coherence directory eviction mechanisms for unmodified copies of memory lines in multiprocessor systems
US6934814B2 (en) * 2002-11-05 2005-08-23 Newisys, Inc. Cache coherence directory eviction mechanisms in multiprocessor systems which maintain transaction ordering
US20040088496A1 (en) * 2002-11-05 2004-05-06 Newisys, Inc. A Delaware Corporation Cache coherence directory eviction mechanisms in multiprocessor systems
WO2004068361A1 (fr) * 2003-01-27 2004-08-12 Fujitsu Limited Dispositif de commande de memorisation, dispositif de commande de cache de donnees, unite centrale, procede de commande de dispositif de memorisation, procede de commande de cache de donnees et procede de commande de cache
US20050210204A1 (en) * 2003-01-27 2005-09-22 Fujitsu Limited Memory control device, data cache control device, central processing device, storage device control method, data cache control method, and cache control method
US9558116B2 (en) 2003-04-09 2017-01-31 Intellectual Ventures Holding 81 Llc Coherence de-coupling buffer
US10540283B2 (en) 2003-04-09 2020-01-21 Facebook, Inc. Coherence de-coupling buffer
US20130046961A1 (en) * 2011-08-15 2013-02-21 Alexander Rabinovitch Speculative memory write in a pipelined processor
CN103728897B (zh) * 2012-10-12 2016-12-21 深圳市金正方科技股份有限公司 一种基于软硬件结合的抗干扰方法及系统
CN103728897A (zh) * 2012-10-12 2014-04-16 深圳市金正方科技股份有限公司 一种基于软硬件结合的抗干扰方法及系统
US9134377B2 (en) * 2013-03-14 2015-09-15 Teradyne, Inc. Method and apparatus for device testing using multiple processing paths
KR20150128757A (ko) * 2013-03-14 2015-11-18 테라다인 인코퍼레이티드 다중 처리 경로를 사용하는 장치 검사 방법 및 장치
US20140281776A1 (en) * 2013-03-14 2014-09-18 Teradyne, Inc. Method and apparatus for device testing using multiple processing paths
KR102126763B1 (ko) 2013-03-14 2020-06-25 테라다인 인코퍼레이티드 다중 처리 경로를 사용하는 장치 검사 방법 및 장치
US10139449B2 (en) 2016-01-26 2018-11-27 Teradyne, Inc. Automatic test system with focused test hardware
US10907426B2 (en) 2018-10-15 2021-02-02 H. Udo Zeidler Apparatus and method for early kick detection and loss of drilling mud in oilwell drilling operations

Also Published As

Publication number Publication date
EP0495165A2 (fr) 1992-07-22
JPH05303492A (ja) 1993-11-16
DE69130858T2 (de) 1999-10-07
DE69130858D1 (de) 1999-03-18
EP0495165B1 (fr) 1999-02-03
EP0495165A3 (fr) 1992-08-12
BR9200054A (pt) 1992-09-08
CA2056715A1 (fr) 1992-07-17
JPH0785222B2 (ja) 1995-09-13

Similar Documents

Publication Publication Date Title
US5257354A (en) System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results
EP0690371B1 (fr) Tampon d'extraction et stockage de données pour l'exécution hors-séquence dans un système de traitement de données
JP3701814B2 (ja) マルチプロセッサ・システム及びそのシステム・シリアル化方法
US5134561A (en) Computer system with logic for writing instruction identifying data into array control lists for precise post-branch recoveries
US7020871B2 (en) Breakpoint method for parallel hardware threads in multithreaded processor
US8539485B2 (en) Polling using reservation mechanism
US5941981A (en) System for using a data history table to select among multiple data prefetch algorithms
JP2937485B2 (ja) スーパースカラプロセッサにおけるトラップを検出して実行する方法及び装置
US5694617A (en) System for prioritizing quiesce requests and recovering from a quiescent state in a multiprocessing system with a milli-mode operation
US5499356A (en) Method and apparatus for a multiprocessor resource lockout instruction
US6098167A (en) Apparatus and method for fast unified interrupt recovery and branch recovery in processors supporting out-of-order execution
US6301655B1 (en) Exception processing in asynchronous processor
US20050066131A1 (en) Apparatus and method for loading data values
KR100242403B1 (ko) 데이타 처리 시스템의 조건 레지스터 데이타 임시 저장 방법 및 시스템
US5003458A (en) Suspended instruction restart processing system based on a checkpoint microprogram address
EP0614138A1 (fr) Processeur de données avec bit pour indiguer le registre attribué le plus récemment
US6148394A (en) Apparatus and method for tracking out of order load instructions to avoid data coherency violations in a processor
US6735688B1 (en) Processor having replay architecture with fast and slow replay paths
US5930491A (en) Identification of related instructions resulting from external to internal translation by use of common ID field for each group
US5469552A (en) Pipelined data processor having combined operand fetch and execution stage to reduce number of pipeline stages and penalty associated with branch instructions
EP0550283A2 (fr) Appel d'actions de restauration de matériel par la voie de bascules d'action
EP0374370B1 (fr) Procédé de mémorisation dans des lignes d'antémémoire non exclusives dans des systèmes multiprocesseurs
US5146569A (en) System for storing restart address of microprogram, determining the validity, and using valid restart address to resume execution upon removal of suspension
US5996085A (en) Concurrent execution of machine context synchronization operations and non-interruptible instructions
US6134645A (en) Instruction completion logic distributed among execution units for improving completion efficiency

Legal Events

Date Code Title Description
AS Assignment

Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, ARMON

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST.;ASSIGNORS:COMFORT, STEVEN T.;LIPTAY, JOHN S.;WEBB, CHARLES F.;REEL/FRAME:005582/0825

Effective date: 19910110

FEPP Fee payment procedure

Free format text: PAYOR NUMBER ASSIGNED (ORIGINAL EVENT CODE: ASPN); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

FPAY Fee payment

Year of fee payment: 4

FEPP Fee payment procedure

Free format text: PAYER NUMBER DE-ASSIGNED (ORIGINAL EVENT CODE: RMPN); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

REMI Maintenance fee reminder mailed
LAPS Lapse for failure to pay maintenance fees
STCH Information on status: patent discontinuation

Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362

FP Lapsed due to failure to pay maintenance fee

Effective date: 20011026